HIDALGO URBAN AIR POLLUTION PILOT BASED ON CAMS DATA
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
HiDALGO urban air pollution pilot based on CAMS data Zoltán Horváth, Bence Liszkai, Ákos Kovács, Tamás Budai and Csaba Tóth Széchenyi István University, Győr, Hungary CAMS 4th General Assembly and User Day Budapest, 16-20 September 2019 HiDALGO – EU founded project #824115
Vision of HiDALGO’s Urban Air Pollution Provide citizens and policy makers with forecast and reanalyses at very high resolution for urban air pollution using science and supercomputing resources through an easy web-interface. 30.09.2019 © HiDALGO 2
Agenda 1. The global challenge: improve urban air quality 2. The HiDALGO digital twin for urban air pollution 3. Demonstration to Győr and Stuttgart 30.09.2019 © HiDALGO 3
Global challenge: improve urban air quality • 3 million deaths attributable to ambient air pollution, by WHO | https://www.who.int/phe/health_topics/outdoorair/databases/en/ • Traffic is emitting >40% of several contaminants (e.g. NO2) 30.09.2019 © HiDALGO 4
Global challenge: improve urban air quality • EC regulates air quality management and allows the use of computational models for reporting (see Directive 2008/50/EC). • EC provides forecasts for air quality from CAMS: | European AQ – Ensemble hourly forecasts and analyses | AQI for every 3 hours, for several cities, one value for the whole town • However, … 30.09.2019 © HiDALGO 5
Global challenge: improve urban air quality • Example: Győr, Hungary. NO2 simulation with 3D geometry | Hot spots occur even when overall city values are OK | High resolution validated simulation is needed → need of CFD & HPC | recent activities are starting within FAIRMODE as well 30.09.2019 © HiDALGO 6
The HiDALGO digital twin for urban air pollution • Background: | H2020: MSO4SC, CoeGSS | Hungarian-ESF-projects: SZE FIEK (GINOP) • HiDALGO – Center of Excellence for HPC and Big Data for Global Challenges | H2020 CoE project, from December 2018 until November 2021 | provides HPC, HPDA infrastructure by experts, and pilot services based on the infrastructure for global challenges | HiDALGO is to do ”the heavy weight lifting for modelling - HPC, HPDA and algorithms - of global challenges and some finale mile runs” | Technical coordinator: HLRS, coordinator: ATOS • Goal of HiDALGO urban air polution (UAP) pilot: develop a service for UAP with very high resolution • Demonstration area: Győr, Hungary (of 130.000 habitants) 30.09.2019 © HiDALGO 7
What is a model based digital twin? • HiDALGO UAP as a digital twin • Digital twin = digital replica of a real physical asset for which | digital image is based on computational simulations of physical models, | connected with sensor measurements to the real asset, | models updated continuously upon measurements, | gives real time answers to questions on the real asset (based on model order reduction) • More details: see the booklet by EU-MATHS-IN on technologies for digital twinning (much) beyond the state-of-the-art 30.09.2019 © HiDALGO 8
HiDALGO digital twin for urban air pollution - goals • Highly accurate simulation of urban air pollution | Real 3D geometry of the city | High resolution mesh: 1 m at street level | Online and real time sensor data from sensor networks (cameras with plate number recognition and low cost AQ) | Traffic emission: from SUMO simulation or statistical data | Weather forecasts from ECMWF | CAMS data for background concentration, local emissions, and long distance emission, all on coarse grid | Highly accurate simulation (CFD) for wind and dispersion • Model order reduction and ensemble modelling for the pollution • Service to be developed, aim: CAMS use case • Traffic management (model predictive control) based on the digital twin 30.09.2019 © HiDALGO 9
The HiDALGO digital twin for urban air pollution • Sensors1: Intelligent camera based sensor network of the traffic | Implementation is ongoing by Adaptive Recognition Hungary (ARH) and Hungarian Public Road Ltd (MK) | Plate number recognition and loop detector data | Generate full trip information, origin-destination matrix | Data will be anonymised and transported real time to SZE directly | SUMO model will be updated real time based on data assimilation 30.09.2019 © HiDALGO 10
The HiDALGO digital twin for urban air pollution • Sensors2: Weather and background pollution data | Weather and pollution sensor data are assimilated into simulation for predictions and reanalyses | Weather (forecast and reanalyses) data are provided by ECMWF through | data exchange and postprocessing via Python scripts (now) | REST API service of ECMWF (to be developed in HiDALGO) | Weather data are used for | boundary conditions for the city wind field computations | advanced physical models (with radiation, humidity, etc; to be developed) | Pollution data are used from the CAMS simulations and observations | background concentration, | local emissions, | long distance emission, all on coarse grid to be developed 30.09.2019 © HiDALGO 11
The HiDALGO digital twin for urban air pollution Overview of workflow Workflow of the initial version (MSO4SC 3DAirQualityPrediction) 30.09.2019 © HiDALGO 12
Module configuration • Input parameters | set in a text file (e.g. see that below for the dispersion module), | some of them edited in the portal GUI (in TOSCA blueprint) | input-output files are standardized (→provides opportunity for changing solvers) #!/bin/bash ENCAS_SAVING_ENABLED="True" # Parameters for the dispersion module ENCAS_OUTPUT_PREFIX="model_result" #### Fluent configurations #### ENCAS_SAVING_PERIOD=10 FLUENT_BINARY="fluent" NUMBER_OF_CORES=2 FLUENT_CUSTOM_COMMAND_LINE_OPTIONS="" STATE_MATRIX_SAVING_ENABLED="True" STATE_MATRIX_OUTPUT_PREFIX="state-matrix" STATE_MATRIX_FIELDS="nox-ppb x-velocity y-velocity z-velocity" #### Simulation and model parameters #### STATE_MATRIX_SAVING_PERIOD=10 SIMULATION_START_TIME="2017-05-10 00:00:00" ITERATION_STEADY_FOR_INITIALIZATION=30 ITERATION_TRANSIENT_PER_TIMESTEP=5 # Monitors (geometry-name field) TIMESTEP_SIZE_SECONDS=60 MONITORS[0]="central_point nox-ppb" NUMBER_OF_TIMESTEPS=30 MONITORS[1]="central_point x-velocity" # No2 concentration calculation ppb = no2.mass.fraction*1e9*46/28 MONITORS[2]="central_point y-velocity" # 20 [ppb]=0.00000001217[no2-mass-fraction] MONITORS[3]="central_point z-velocity" NOX_BACKGROUND_MASS_FRACTION=0.00000001217 MONITORS[4]="side_point nox-ppb" #### Geometry definitions #### MONITORS[5]="surface_2m nitrogen-dioxide" # STL Surface definitions (surface_name stl-path) STL_SURFACES[0]="surface_2m slicer_surface.stl" #Plots (surface1,surface2,surfaceN field min-val max-val) wher min-val and max-val are optional # Point definitions (name x y z) POINTS[0]="central_point 66.40663 69.79992 16.45698" PLOTS[0]="wall_ground,wall_building velocity-magnitude" POINTS[1]="side_point 58.4459 67.79778 16.45698" PLOTS[1]="surface_2m nitrogen-dioxide 0 0.0000001085" PLOTS[2]="surface_2m nox-ppb 0 50" #### Output controls #### PLOTS[3]="surface_2m velocity-magnitude" # Full domain outputs PLOTS_SAVE_PERIOD_TIMESTAMPS=1 CASE_AND_DATA_SAVING_ENABLED="True" CASE_AND_DATA_OUTPUT_PREFIX="model_result" PLOTS_RESOLUTION="1920x1080" CASE_AND_DATA_SAVING_PERIOD=10 30.09.2019 © HiDALGO 13
The HiDALGO digital twin for urban air pollution • Demonstration and validation to Győr | Area: 4 km x 4 km x 0.8 km | Mesh: 800.000 octree cells; meshsize: from 2 m (street) to 50 m | Computation time: 1/3 of simulation time on 16 cores 30.09.2019 © HiDALGO 14
The HiDALGO digital twin for urban air pollution Dispersion computation • ANSYS Fluent module for wind and dispersion simulation (transient Navier-Stokes with turbulence modelling, diffusion-advection-reaction of NOx, O3; transient boundary conditions) • Experiments with an open source, free software as well (e.g. OpenFOAM) 30.09.2019 © HiDALGO 15
Application to Stuttgart Application to Stuttgart (test of the HiDALGO urban air pollution pilot) | All preprocessing steps took 5 person days | 3D geometry is generated from Open Street Map - Video | Meshing is done via in-house octree-mesher | Traffic is simulated with SUMO based on synthetic data | Weather data are from ECMWF forecast to 2019-05-25 | Northern wind of 2 m/s, in most of the day 30.09.2019 © HiDALGO 16
Further applications – illustrations to Stuttgart 30.09.2019 © HiDALGO 17
Towards the digital twin Model order reduction (MOR) of the air flow computation and dispersion modelling (ongoing) | Snapshot matrix compilation is solved from workflow | SVD for one use case (with transient wind boundary conditions for the Navier-Stokes equations (Re=10^9) and transient emission): s. values drops 2 magnitude with < 20 s. bases vectors) → good starting point for the MOR | Note: HPC is used at composition of the reduced (i.e. computationally cheap) models only; at production only cheap models will run. 30.09.2019 © HiDALGO 18
The HiDALGO digital twin for urban air pollution • Usability: run the simulation on HPC from simple, web based portal → HPC is reachable for policy makers easily! • Now the MathSO portal operates → online demonstration! https://youtu.be/RV1Tg7-Rl1c demonstration video 30.09.2019 © HiDALGO 19
Conclusions and further steps Done: operational simulation infrastructure of urban air pollution with | CFD for dispersion | HPC (use of supercomputers) | Easy-to-use web based user interface | Fast preprocessing, enabled by developed tools Next steps | Physical model to be developed | Implementation of model order reduction for faster simulation | Coupling with CAMS data | New indicators to be worked out according to high resolution | More requirements gathering 30.09.2019 © HiDALGO 20
THANK YOU ! QUESTIONS ? Prof. Zoltán Horváth Széchenyi István University Egyetem tér 1. 9026 Győr, Hungary Phone: +36-96-613657 Email: horvathz@math.sze.hu 30.09.2019 HiDALGO – EU founded project #824115 21
You can also read